Aspect-level sentiment classification based on aspect-oriented information and inter-aspect relations

Abstract

Aspect-level sentiment classification aims to determine the sentiment polarity of a given target aspect in a sentence. To solve the problem of ignored the impact of noise in sentences and sentiment relations between different aspects on the sentiment classification performance of models in the current studies, this paper proposes an aspect-oriented syntactic dependency graph and an inter-aspect dependency tree. Based on it, an interactive graph attention network model is proposed to extract sentiment features of the target aspect by exploiting aspect-oriented and inter-aspect information. Experimental results on SemEval-2014 and Twiter datasets show that the sentiment classification ability of the model is superior to the baseline models, and the accuracy of sentiment classification on restaurant review dataset (Rest14) reach 83.36%.